Beyond Gaming - NVIDIA GeForce GTX 280 1GB GDDR3

A New Name for a New Generation

While NVIDIA's GeForce 9 series further optimized the unified shader architecture that debuted with the GeForce 8800 GTX (G80), there were no significant changes in the architecture. The flagship member of the series, the GeForce 9800 GTX packed the same number of stream processors as the original G80, with the die shrink to 65nm the biggest physical change. The other was an increase in transistors due to the integration of the much enhanced VP2 video processor engine for hardware accelerated HD playback into its G92 core.

Therefore, given the short life-cycle of the average GPU, it's about time NVIDIA came up with its next generation architecture. And the graphics industry leader has not only delivered an improved unified architecture, it has also revamped its naming scheme. Partly, this is due to the company reaching a natural limit to its four digit scheme with the GeForce 9000 series, while its products and their numerous prefixes were confusing some consumers. ATI has also recently simplified its product lineup starting with Radeon HD 3000 series so this move by NVIDIA to follow suit is not unexpected.

The new series, now simply known as the GeForce GTX 200, is based on a new unified shader architecture, though compared to the previous generation, this is an evolutionary step. Physically, the GTX 200 GPUs are the largest and most complex graphics processors designed by NVIDIA and are built by TSMC on a 65nm manufacturing process (still the same process technology unfortunately). With a transistor count of 1.4 billion, it is roughly double the 681 million on the first generation G80 architecture. This stat alone is staggering as it would be a very expensive part for NVIDIA and a possible concern of heat output.

NVIDIA has kept with the underlying logic and architecture but enhanced the core in a few ways, like increasing the number of stream processors, improved texture performance, a 512-bit memory interface, and power management enhancements to cope with the demands of modern 3D graphics. On the other hand, the new GPUs are DirectX 10 and Open GL 2.1 compliant and unlike rival ATI, the company will not be formally moving to DirectX 10.1 compliancy just yet, as NVIDIA believes the new features in 10.1 are minor and incremental while some are already implemented in hardware. With only one game, Assassin's Creed that used to support DirectX 10.1, (a recent subsequent patch removed it, generating some controversy), and it's hard to see much progress in this area. Especially when you consider that DirectX 11 slated for next year isn't that far away, so it's quite logical for some developers to jump an incremental version and support the next iteration when it's ripe.

Echoing NVIDIA's recent theme of advancing the usage of GPUs beyond its traditional domains in 3D applications like games, these new GPUs have another side that gaming enthusiasts usually neglect, which happens to be general computing (GPGPU) and like the GeForce 8 and 9 series, these GPUs will support NVIDIA's CUDA technology, with each stream processor executing an instruction thread like a massively parallel computer. Before we proceed to the architectural details, here's a table listing the specifications of the two models available at this moment, the GeForce GTX 280 and 260 next to existing high-end GPUs.

GeForce GTX 200 Architectural Enhancements

Below is an overview of the graphics architecture on the GTX 200 series of GPUs. It's not that much different from NVIDIA's existing first generation. Basically, there's a thread dispatch logic unit, along with setup/raster units that takes the given instructions and assigns them to the texture processing clusters, each of which has a set of general purpose processors which can execute all types of instruction threads, whether they are pixel, vertex, geometry or compute. The ROPs and the memory interface units make up the rest of this architecture.

As you can see, the performance of this thread scheduler will directly determine whether the GPU will be operating effectively or whether there are idle processors. According to NVIDIA, the thread scheduler can support over 30,000 threads simultaneously, an increase from the appropriately 12,000 on the GeForce 8/9 series, while also improving in efficiency to reach almost 100%. The GPU can also switch threads to process with very low latency.

The More (Stream Processors), the Merrier

Just like how the CPU manufacturers engage in their race to squeeze more cores into a single die, there is a similar process going on for the GPU. The difference is that the general purpose processing cores that form the basis of the unified shader architecture underlying both ATI and NVIDIA's present generation of GPUs number in the hundreds, though obviously they are not as complex as that found in CPUs.

For the case of the original first generation GeForce 8800 GTX (G80), there are 128 streaming processors. This number has been almost doubled, to 240 for the flagship GTX 280 GPU and 192 for the performance model, the GTX 260. This is achieved by tweaking the prior arrangement of the stream processors, with NVIDIA engineers increasing the number of clusters of such processors (TPCs) while at the same time, increasing the number of multi-processors (SMs) in each cluster. Of course, you'll find within each streaming multi-processor group the necessary supporting infrastructure like Texture Mapping Units (TMUs) and local memory caches.

Obviously, the increase in processor cores and thread scheduling has implications beyond graphics. NVIDIA has been rather vocal in stressing the general purpose computing prowess of its GPUs and the same hardware on the GTX 200 series can be converted for parallel computing with the appropriate software (in this case, CUDA). Applications that used to be the domain of the CPU can now be ported to run on the GPU, with significant improvements in performance. These include video transcoding, distributed computing, financial modeling and scientific simulations.

Related to this GPGPU focus, we also find NVIDIA enabling double precision support to the GTX 200 architecture, which is important for high performance computing that requires a high degree of mathematical precision. This is done by adding double precision 64-bit floating math units to the mix (a total of 30), all of which are IEEE754R floating point specification compliant. Accordingly, NVIDIA claims that the overall double precision performance of a GeForce GTX 200 GPU is around the level of an eight-core Xeon CPU. It's also the first NVIDIA GPU to have such a feature.

More Memory, More Bandwidth

While ATI high-end GPUs have been on a 512-bit memory interface for some time now, NVIDIA's equivalent GPUs have been relying on a 384-bit bus at best. This is set to change with the GTX 280, as these GPUs have now moved to adopting a 512-bit memory interface (8 x 64-bit memory interface units). The GTX 260 meanwhile gets a smaller upgrade to 448-bit (7 x 64-bit). NVIDIA also claims to have modified the paths in the memory controllers to allow for higher memory speed and the GTX 280 already runs at a high stock speed of 2214MHz DDR (and we have in fact reached higher speeds during overclocking). Improvements in compression and caching algorithms have also enabled better performance at higher resolutions.

Along with the memory interface boost, the total amount of memory on the GTX 200 series has also gone up. The GTX 280 comes with 1GB of frame buffer and the GTX 260 has 896MB. This should improve the GPU's performance in the latest games with anti-aliasing at higher resolutions, especially for newer games that use deferred shading, which consume more memory.

Internally, NVIDIA has also doubled the file size of the local registers within each SM unit from that on the GeForce 8/9 series. This would reduce instances where large, complex shaders were too big for the registers, necessitating a slower and more inefficient swap to memory. Newer game engines are more likely to benefit from this, as they tend to use more complex shaders.

Texture Filtering and ROP Improvements

Previously, the GeForce 9800 GTX can address and filter up to 64 pixels per clock. The GeForce GTX 200 GPUs bring this number up to 80 pixels per clock for the high-end model, while providing an equal balance between texture addressing and filtering. A more efficient scheduler enables the new GPUs to approach closer to its theoretical peak limits compared to the GeForce 9 series.

The texture hardware count has remained at the same proportion as before but NVIDIA has increased the number of shaders (by having 1 more SM in each TPC), thereby leading to a higher shader to texture ratio. This is done in order to create a more balanced GPU to respond to the needs of modern games and applications, which nowadays are shifting more to complex shaders.

In terms of ROP (Raster Operations Processors) hardware, the larger number of TPCs naturally requires a greater number of ROPs for a better balance. Hence, there are now 8 partitions of ROPs on the GTX 200 compared to 6 on the GeForce 9800 GTX. Additionally, ROP frame buffer blending for pixels using 8-bit unsigned integer data format can now be done at twice the speed as before.

Geometry shading and stream out performance has also received an upgrade. NVIDIA has boosted the number of internal output buffer structures by up to six times for the GTX 200. This was done after feedback was received that the present quantity on the GeForce 9 series was inadequate for certain applications.

Using the Right Amount of Power

As one would expect from its number of transistors, the GTX 200 GPUs can consume a staggering amount of power. NVIDIA itself states the maximum TDP of the GTX 280 in full 3D mode is up to 236W. Therefore, NVIDIA has been at pains to introduce a more aggressive dynamic power management architecture for the GTX 200.

Besides supporting NVIDIA's new Hybrid SLI technology, which allows one to turn off the discrete GPU completely (hence drawing 0W of power) in favor of the integrated GPU on the latest NVIDIA nForce motherboard, the GTX 200 can effectively 'turn off' sections of the GPU thanks to its clock-gating circuitry. According to NVIDIA, at idle, the GTX 200 consumes around 25W of power, going up to 35W when playing back HD videos (e.g. using the PureVideo HD capable VP2). Such clock and voltage adjustments are done automatically internally and they are very responsive (in milliseconds) to the utilization rate of the GPU so as to achieve maximum power savings while being able to perform at its peak when necessary.

Using the latest version of GPU-Z utility, we managed to catch a snapshot of the clock speeds on the GeForce GTX 280 ramping up from its idle state when starting 3DMark Vantage. Of course, we aren't completely sure about the accuracy of GPU-Z's readings, given how new the GTX 280 is, but the results seem fairly reasonable.

In case you were keeping count of all that we have mentioned (and some minor ones that we have left out), below is NVIDIA quantifying the improvements that have been made from the GeForce 8800 GTX to the GeForce GTX 280. (Of course, some of the information here like FB bandwidth and PCI Express for example, are moot as these are theoretical numbers that are unlikely to be reached in both cases.)

The NVIDIA GeForce GTX 280 1GB GDDR3

Clad in black with NVIDIA's logo embossed on its heatsink enclosure, the reference board that we received from NVIDIA seems to have gone through a lot, with its scratched surface. But of course, we couldn't care less and quickly plugged it into our test system for a run.

Just to repeat again, the core is at its stated default of 602MHz, while the GDDR3 memory runs at 2214MHz. Its 240 stream processors meanwhile are at 1296MHz. The GeForce GTX 280 has a 512-bit memory interface, with 1GB of memory made up of 16 Hynix memory modules (rated at 0.8ns). There's also an external chip to handle the display I/O, presumably because the core is already so huge that it would be pushing it to squeeze that portion of silicon within. This same approach was also used on the original G80 core.

When running, the smart fan is relatively quiet and we couldn't distinguish it from the heatsink of the CPU. However, this was at around 40% fan speed. When we manually pushed it to 100% in NVIDIA's Control Panel, it was quite loud so if you're planning to do that for overclocking purposes, take note of the potential din. The board itself is rated by NVIDIA to be operating within its thermal threshold as long as it's under 105 degrees Celsius and according to them, it should be around 80 degrees under typical conditions.

Despite its ability to adjust clock speeds dynamically in order to conserve energy during idle periods, you'll still need quite a decent PSU for a single GTX 200 GPU based board. Besides having the right connectors (6-pin and 8-pin for the GTX 280), the PSU should be able to provide at least 40A on the 12V rail for the GTX 280 and 36A for the GTX 260. We'll be looking at some numbers from actual testing of the power consumption of the GTX 280 later.

Test Setup

We had two separate test systems for the GeForce GTX 280, one on Windows XP SP2 and the other using Windows Vista SP1. This was done in order to run some of the newer, more extreme benchmarks that required DirectX 10, while keeping the older system as a reference point for some of the results. In the future, we should be moving over completely to a newer, Windows Vista based test system but for the moment, here are the configurations that we used:

Windows XP SP2 Test System

• Intel Core 2 Duo E6700 (2.66GHz)
• Intel D975XBX 'Bad Axe' motherboard
• 2 x 1GB DDR2-800 Kingston HyperX memory in dual channel mode
• Seagate 7200.7 80GB SATA hard drive
• Windows XP Professional with Service Pack 2 and DirectX 9.0c

Windows Vista SP1 Test System

• Intel Core 2 Extreme QX6850 (3.00GHz)
• Gigabyte X38T-DQ6 motherboard
• 2 x 1GB DDR3-1333 memory in dual channel mode
• Seagate 7200.10 200GB SATA hard drive
• Windows Vista Ultimate with SP1

We compared the GeForce GTX 280 against existing high-end GPUs from both ATI and NVIDIA, including the GeForce 9800 GX2, GeForce 9800 GTX and the Radeon HD 3870 X 2 and a CrossFireX pair of Radeon HD 3870 X2. As for the graphics drivers, the GeForce GTX 280 was tested using the latest ForceWare 177.34 from NVIDIA for both the Windows XP and Vista versions. The other NVIDIA cards were running ForceWare 175.16 while the ATI cards were on Catalyst 8.5.

The following benchmarks were tested using their built-in time demo or benchmarking tools:

For Windows XP:

• Futuremark 3DMark06 (ver 110)
• Company of Heroes (ver 1.3)
• Crysis (ver 1.1)
• Unreal Tournament 3 (ver 1.1)

For Windows Vista

• Futuremark 3DMark Vantage (ver 101)
• Crysis (ver 1.21)

Win XP Results - 3DMark06 (ver 110)

The GeForce GTX 280 may be the latest from NVIDIA but the company's dual GPU solution, the GeForce 9800 GX2 immediately put the dampeners on the hype by outdoing the new card in 3DMark06. ATI's competing Radeon HD 3870 X2 too was up to the task, edging out the GTX 280 when anti-aliasing was disabled and roughly equal to it when it was enabled. To be fair though, the GTX 280 did manage a significant performance boost over the previous single GPU champion, the GeForce 9800 GTX by a minimum of 6% and up to 40% at the more extreme settings.

Win XP Results - Company of Heroes & Unreal Tournament 3

Company of Heroes, the first real game we tested helped to validate the GTX 280, as it managed to beat both the dual-GPU cards. The difference however was quite modest at around 5 - 7% between the GTX 280 and the GeForce 9800 GX2. Between the GeForce 9800 GTX and the GTX 280, there was a significant margin of up to 34%.

Unreal Tournament 3 was not as convincing for the new arrival, as the GeForce 9800 GX2 continued to argue its case, especially at the highest resolution tested of 1920 x 1440. Again, looking at the GeForce GTX 280 and the GeForce 9800 GTX, the second generation card did have a massive 50% lead at 1920 x 1440.

Win XP Results - Crysis (ver 1.1)

It was in Crysis that the GeForce GTX 280 arguably had its finest hour so far. At long last, we finally saw a graphics card producing a playable fps of around 30 for the game at High Settings and at 1920 x 1440. Compared to the other high-end cards here, there was no doubt that the GTX 280 deserved its next generation billing. Even at less extreme settings, the GTX 280 was mostly in the lead and only once did the GeForce 9800 GX2 come close.

Win Vista Results - 3DMark Vantage

Now that we saw how the GeForce GTX 280 performed in Windows XP benchmarks, how about Windows Vista, especially in the new 3DMark Vantage benchmark? This new benchmark came with four performance presets, each representing certain resolutions and graphical settings. We tested the cards here in Performance, High and Extreme settings.

The GeForce GTX 280 held its own here, taking the crown for the High and Extreme presets. This was in spite of competition from a CrossFireX setup of two Radeon HD 3870 X2 and the usual bunch of high-end cards that we have seen earlier. At the Extreme preset, the GTX 280 was twice as fast as the GeForce 9800 GTX.

Win Vista Results - Crysis (ver 1.21)

After what we saw of the GTX 280 in the Windows XP test, we decided to notch it up for the Windows Vista run. This meant changing the settings to Very High, which led to quite a drastic drop in frame rates for all the cards. The dual-GPU GeForce 9800 GX2 turned out to handle it better than the rest and handily beat the GTX 280. After all, the GeForce 9800 GX2 does actually have higher clocks and more processing units than the single GTX 280, so this sort of outcome is no stranger to us.

The GTX 280 however managed to secure second place while the two Radeon HD 3870 X2 in CrossFireX seemed to have some issues here, with the 4 GPUs solution faring similar or worse than a single Radeon HD 3870 X2. With anti-aliasing enabled, the GTX 280 showed its potential while for some reason, our GeForce 9800 GX2 and GTX kept crashing and hence were unable to finish the benchmark.

Temperature

The GeForce GTX 280 performed much like NVIDIA claimed as we recorded a core temperature of around 72 degrees Celsius during our temperature test. This was slightly better than the GeForce 9800 GX2 and comparable to the other high-end GPUs. Note that this was conducted with the GTX 280 smart fan in automatic mode, so it wasn't noisy.

Power Consumption

With a power meter, we could measure the power draw of the entire system and the results below are what we got for the various high-end cards, on Windows Vista running 3DMark Vantage Performance preset. A 1200W Thermaltake PSU was used for this. While our test was unable to isolate the specific power draw of the GeForce GTX 280, the idle power draw of the entire system with that card was lower than all the other cards here, which showed that NVIDIA had indeed improved the dynamic power management on the GTX 280. At peak however, the GTX 280 was as power hungry as the high-end GPUs nowadays and our readings more or less correlate with what we know about this card (236W maximum TDP).

Overclocking

Even though its default clock speeds seemed decently high, especially the 2214MHz stock GDDR3 memory clock, the GTX 280 still had some allowance for overclocking. We manually set the fan speed to maximum, which meant it was quite noisy and then started pushing up the bars. The GTX 280 could reach to 700MHz for the core, an improvement of around 100MHz and a surprisingly high 2650MHz DDR for the memory. Based on that clock speed, there was a 18% increase in 3DMark Vantage scores using the Extreme profile, a rather nice performance boost for something that was easily done through NVIDIA's Control Panel.

Conclusion

From a technical point of view, the architecture on the new GTX 200 GPUs is not significantly different from the original GeForce 8800 GTX. Considering that the GeForce 8800 GTX represented a major shift from distinct pixel and shaders to unified shaders, it would be quite unfair to judge the GTX 200 on that basis. Perhaps in the near future, we may see another new architecture (ray tracing hybrid maybe?) but for the moment, the GTX 200 adds even more hardware, while having quite a few internal tweaks to gear it for modern games with their increasingly complex shaders.

With its stream processors increased to 240, the GeForce GTX 280 is the new flagship GPU from NVIDIA and the largest. From that alone, one can expect a performance boost, all things being equal. Of course, NVIDIA did have other enhancements to improve it further, including an extensive list of optimizations, like an increased output buffer, larger registers, increased memory bandwidth to 512-bit wide, a 1GB frame buffer size, double precision floating-point support and so on. Some of these are not easily reflected in raw performance, though it could make the GTX 280 more useful in general computing.

This seems to be NVIDIA's focus in recent times, with the company emphasizing on the growing importance of its GPUs in these parallel computing tasks. Like the GeForce 8/9 series, the new GTX 280 are CUDA capable and NVIDIA's purchase of Ageia also means that PhysX will be ported over to take advantage of these GPUs in physics processing. In short, besides its graphics prowess, getting a modern NVIDIA GPU will have such additional benefits and features and you will definitely see such capabilities on the GTX 200.

When it comes to gaming performance, the GeForce GTX 280 easily justified its billing and improved over the former single GPU champion from NVIDIA, the GeForce 9800 GTX. However, when it came to some of the newer dual-GPU single-card comparisons, the newcomer has some ground to make up. Depending on the games and settings, the GeForce 9800 GX2 could be more than a match for the GTX 280 and similarly, would the Radeon HD 3870 X2 (the latter being much more 'affordable' and thus easily a consideration for many). The power management improvements alone however can be quite significant over the previous high-end GPUs and Hybrid SLI is also possible for the least power draw.

We had a peek at NVIDIA's presentation slides for the GTX 200 and in the new world order, there was no place for the GeForce 9800 GX2 at the top. It does seem likely that the GeForce 9800 GX2 will suffer the same fate as the previous GX2, replaced by a newer card. For all this, NVIDIA is asking quite the premium for its new high-end flagship, with the GTX 280 priced at a whopping US$649 at launch, far beyond any existing high-end GPUs now. By contrast, the GTX 260 will be available for US$399, which once you consider the likely performance delta between the two, looks like much better value. If getting the best is your main prerogative then the GTX 280 is exactly that, but you'll have to pay dearly for it. The key to really take advantage of the GTX 280 is cutting edge games at extreme quality settings as well as GP-GPU computing tasks, so keep these in mind and you won't be disappointed. For all other purposes and needs, the GTX 280 would easily be an overkill.

Stay tuned as we will shortly bring you a review of the GTX 260 as well for your full consideration of the new series.

Our articles may contain affiliate links. If you buy through these links, we may earn a small commission.